Description
Applications are increasingly deployed in the cloud, but limiting the trust required in
the cloud-provider remains challenging. Existing solutions that build upon TEEs only
allow the integrity of the environment to be attested at launch-time and to protect data
confidentiality. Protecting the integrity of memory is still, at best, a partially solved
problem. This thesis proposes a redundancy-based approach. It executes a singlethreaded
containerized application a second time at a different cloud-provider in a
different execution environment and compares the resulting behavior to detect integrity
violations. The approach requires reproducible executions. This thesis investigates
the extent to which such reproducibility can be achieved across heterogeneous Linuxbased
execution environments. We implement a proof-of-concept and evaluate how
different observation levels and execution environment differences affect reproducibility.
We identify incompatibilities between environments and analyze ways to remove or
circumvent them. Our results show that reproducible execution is feasible, but only for
compatible execution environment pairs with respect to the chosen observation level.
Overall, this thesis contributes a design and an experimental analysis demonstrating
that reproducible execution across heterogeneous Linux-based environments is feasible,
but depends critically on the chosen environment pair and observation level.
|
Thesis topic in co-operation with the Fraunhofer Institute for Applied and Integrated Security AISEC, Garching